Neural network models based on clinical characteristics for predicting immunotherapy efficacy in small cell lung cancer

Background

Immunotherapy combined with chemotherapy has been approved as first-line therapy for small cell lung cancer (SCLC) due to the survival benefit in randomized controlled trials. However, predicting its efficacy remains a challenge in the absence of currently available biomarkers.

Methods

A total of 140 individuals with SCLC who received immunotherapy were evaluated retrospectively. These patients were split into two distinct cohorts, the discovery cohort (80% of the total, n = 112) and the validation cohort (20% of the total, n = 28). The objective response rate (ORR), disease control rate (DCR), and responder (progression-free survival [PFS] > 6 months) were all predicted using neural networks.

Results

We developed predictive models for three clinical outcomes. ORR scored 0.8964 area under the receiver operating characteristic curve (AUC) in the discovery cohort and 0.8421 AUC in the validation cohort. DCR model had AUC of 0.8618 in the discovery cohort and AUC of 0.7396 in the validation cohort. The responder model had AUC of 0.8116 in the discovery cohort and AUC of 0.7041 in the validation cohort. The models were then compressed into a doctor-friendly tool.

Conclusion

These neural network-based models, which are based on routine clinical characteristics, accurately predict the efficacy of immunotherapy in patients with SCLC, particularly in terms of ORR.